Back Propagation Neural Network Model Applied In The Analysis Of Relative Factors And Predicition Of The Aggravation Severity Progress Of The Hand-foot-mouth Disease

ObjectivesThis study is conducted to explore the application of Back Propagation Neural Networks model (BPNN) in the analysis of relative factors and prediction of the aggravation severity progress of the hand-foot-mouth disease, being aimed to provide the reference for the clinical diagnosis and prediction of the aggravation severity progress of hand-foot-mouth disease, and lay the foundation for the clinical diagnosis and epidemiology of hand-foot-mouth disease.MethodsBased on status quo investigation data of HFMD, clinical data of344cases of HFMD children were cluster collected during2013.04-2013.06as questionnaire investigation in Zhengzhou Children’s Hospital. The neural network toolbox of MATLAB7.0is adopted to construct the BPNN. Then the mean impact values of relative factors of HFMD aggravation are calculated and sorted by their absolute values. The result is compared with that acquired by logistic regression analysis. Finally, the key relative factors influencing HFMD aggravation are drawn on balance. In addition, we drawed the influential MIV-weighted factors and normalized them. According to the onset to complete data intensive process of clooecting cases, we seek the law of further development of the level of the combination of factors changing with the course of disease.Results1. The results of single-factor logistic regression analysis are: poor spirits, high sugar levels, stiff neck, easily offended (jumpy), somnolence, hand and foot quivering, vomiting, muscle weakness, maximum body temperature≥39℃,and white blood cell count≥15×109/L. The differences of these nine factors between severe and mild cases are statistically significant. Results of multi-factor logistic regression analysis, seven factors enter the model of risk factors influencing the HFMD aggravation, i.e., easily offended (jumpy), hand and foot quivering, somnolence, vomiting, poor spirits and white blood cell count≥15×109/L stiff neck.2. The model achieves the testing dataset and100%for the training dataset, for the classification accuracy of greater than90%, which can match well with the collected data.3. The structure of the BPNN model is27â†'8â†'1.The first ten risk factors influencing the HFMD aggravation acquired(MIV absolute value)are:easily offended (0.4614), poor spirits (0.3050), hand and foot quivering (0.1019),vomiting(0.0912), days of high fever≥3days(0.0711), stiff neck(0.0461), white blood cell count≥15×109/L(0.046), somnolence(0.028), high sugar levels (0.015), abnormalities of respiratory rhythm(0.012).4. Comparison of results of BPNN model and logistic regression analysis shows that the orders of importance of relative factors by two methods are coincidence basically. Moreover, there exsits interactions with high fever≥3days, poor spirits and white blood cell count≥15×109/L, it is found the factor of days of high fever≥3days is an important covariant.5. In the prediction of the aggravation severity progress of the hand foot mouth disease, The level of the combination of factors, which is calculated based on MIV-weighted factors, rises significantly before the days of the aggravation severity progress, the process of rising slightly in the day of severe and severe day, reaches its pesk on the intensive transformation day, and then starts to go down.ConclusionsThe BPNN model deals well with the complicated relationships among relative factors and gives severe warning to the new HFMD cases. The BPNN model can be used for the analysis of the relative factors influencing the HFMD aggravation.